|
Activity Number:
|
40
- Survey Weighting, Imputation, and Estimation
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 3, 2020 : 10:00 AM to 2:00 PM
|
|
Sponsor:
|
Government Statistics Section
|
|
Abstract #312419
|
|
|
Title:
|
A Bayesian Record Linkage Procedure That Adjusts for Variables in One File
|
|
Author(s):
|
Mingyang Shan* and Roee Gutman
|
|
Companies:
|
Eli Lilly and Company and Brown University
|
|
Keywords:
|
Record Linkage;
Multiple Imputation;
Bayesian Inference;
Missing Data
|
|
Abstract:
|
Existing record linkage procedures often rely on comparisons between semi-identifying information that exist in all data sources to merge files together. Subsequent analysis on linked datasets assumes that the linkage is non-informative, which states that the linkage and post-linkage analysis are conditionally independent given the linking information. In scenarios where linking information is limited or the proportion of linkage error is high, this assumption is often violated, which can lead to biased point and interval estimates using linked data. We propose a Bayesian record linkage procedure that incorporates the relationship between variables that are exclusive to one file, in addition to comparisons between linking variables that exist in both files. We demonstrate that this method provides improved linkage accuracy when linking variables are limited or prone to error and improves post-linkage inference when the non-informative linkage assumption may be violated.
|
Authors who are presenting talks have a * after their name.
